wojtek@wojtek.from.pl asks:
>> Dump starts by getting info about any specified filesystem, using file
>> metadata for its database of inodes to dump.  Dump does make sure to
>> handle sparse files correctly.  (Files with holes like some db files.)
>> Dump uses the fts functions to walk the inode list.
>> 
>> Now you can specify as your dump file /dev/<rawdisk> if it's really
>> important to you.
>it is not important but it would not pollute buffer cache...
>
>wouldn't tar have the same speed?

I don't know.

When using dumping on multiple filesystems it does some kind of
multi-process magic.  I don't know how this affects tar.  I've never
had to optimize my tape image creation for speed.  

One nice thing about dump and restore is that you can cd into the
restored tape and find a file, pluck it out and restore only that
file, since dump puts some superblock-like information at the head of
the tape.  Tar I think is more of a monolithic archiving format.

Regards,
James